Bulat, On 7/2/06, Bulat Ziganshin wrote:

can you please describe that "code assistance" means?

You may know it as code completion, completion assistance or code suggestion. It is a much used feature on IDEs: just type your code, hit a special key combination and the program will be able to guess what you mean depending on the context and suggest some options for you. You could try with EclipseFP, just hit Ctrl-Space. Regards, Thiago Arrais

Bulat, Thanks for the suggestions (by the way, very good article on IO). Some comments below. On 7/3/06, Bulat Ziganshin wrote:

- background compilation with errors highlighting

There is already some support for background compilation using GHC. Error highlighting doesn't still work as I'd like to, but is there too.

- syntax highlighting, including highlighting of bracket's pair

There is already some of this too, although it may not be exactly what you have in mind.

- ability to "fall in" definition of identifier under cursor

Being worked on. Should be one of the leading features on the next release, together with some more code completion.

- automatic management of import lists

Being worked on, together with code completion. I would like some more detail here, though.

- auto-indenting with user-tuned style

This one seems very challenging and interesting, not a must-have for me but a very attractive feature. I have purposely left some items behind, just because resources are limited and we need to focus on some more valuable features. Do you think the ones I selected are the most valuable ones? Would you mind to delve into some more detail on the topics above on the EclipseFP mailing list? I am particularly interested on these four (in order of priority): - ability to "fall in" definition of identifier under cursor - automatic management of import lists - syntax highlighting - auto-indenting with user-tuned style You can subscribe to our mailing list on http://lists.sourceforge.net/mailman/listinfo/eclipsefp-develop

Are you use GHC library?

Not yet, but it may turn to be a very wise decision to make in the future. EclipseFP is being written in Java, I wonder how the GHC library would be accessed on such a environment. Need to take a look at the paper. Cheers, Thiago Arrais

Hello Thiago, Monday, July 3, 2006, 4:18:01 PM, you wrote:

...

- background compilation with errors highlighting

There is already some support for background compilation using GHC. Error highlighting doesn't still work as I'd like to, but is there too.

now i'm using editor with just syntax highlighting and therefore i wrote about things those lacking are most important for me. first is the compilation. after i've compiled program i don't like to find files, go to specified line just because i'm not Ceasar and can't think about files/lines and program bugs simultaneously. why the error highlighting is a problem? GHC even supports -ferror-spans option in which emacs-style error message is printed (i.e. it includes the exact range of lines/chars what contain error) the next problem is slowness of GHC compilation. There are whole range of methods to deal with it: 1) background compilation, as you already implemented 2) fast "precompilation" with hugs and reporting bugs it encountered and then "true" compilation with ghc - of course, for this sort of scenario the program should be compatible with both compilers 3) "wide" compilation scheme - on first step compile all modules just searching for syntax errors (which should be much faster) and on the second step perform full compilation 4) using GHCi - it's a 2-3 times faster than GHC and moreover you will drop the startup time and the linking time. This also can be combined with GHC compilation - after module has been successfully compiled with GHCi and programmer continues to further develop his code, GHC can be invoked in background to recompile the modules and make them faster work / faster load in next GHCi session 5) using the GHC library to find simple errors just while editing so compilation in most cases will be successful this list contains all ideas i can recall. i personally like the last variant. of course, the compilation should still be performed in background, it just be called much less often and programmer will not wait for its completion. on the other side, for debugging, 4th variant will be more interesting i also mentioned tight integration with WinHugs. are you seen WinHugs 2006 version? are you able to see it? look at least at http://haskell.org/haskellwiki/WinHugs (Win)Hugs remains the fastest way to debug haskell programs and new WinHugs adds new pretty look and usability features to this story. one especially important one is ability to position editor just at the line containing error. It's already supported for Vim and i will be glad to see the same support for Eclipse. May be it just need adding instructions about setting up editor's call (for Vim it is "gvim.exe --remote-silent +%d %s") The almost complete support for WinHugs should include just two facilities: 1) save all, run modified files through the CPP, load module specified 2) same as above plus run the function specified (running modules through C preprocessor required because WinHugs can't preprocess modules and developing any serious program as compatible both with Hugs and GHC is impossible without conditional compilation)

...

- syntax highlighting, including highlighting of bracket's pair

There is already some of this too, although it may not be exactly what you have in mind.

i'm spoiled with vim-like one in FAR :)

...

- ability to "fall in" definition of identifier under cursor

Being worked on. Should be one of the leading features on the next release, together with some more code completion.

i does it on Vim, using modified hasktags (original one finds only functions with explicit signatures). if Eclipse supports 'tags' file, it should be relatively easy, at least for global identifiers (although field names and class methods will be out of luck in this case). what set of indexed identifiers you are plan to support?

...

- automatic management of import lists

Being worked on, together with code completion. I would like some more detail here, though.

if compiler complains about missing identifiers, the plugin should find module where it defined and add this module name (or this identifier if module already imported with just some identifiers) to import list. i personally keep import list splitted into two parts - global and local modules of current project, each list sorted alphabetically it will be also great to delete from import list unused modules/identifiers, but that is very low on my priority list

...

- auto-indenting with user-tuned style

This one seems very challenging and interesting, not a must-have for me but a very attractive feature.

You can start with indenting after lines with "special" words (such as if, when and so on). The catch is what in Haskell many "control structures" are finished on the same line they are started or just partially applied and passed as parameters to some other functions :) I personally want to see as "control structures" many functions i defined myself. It's also true for syntax highlighting - my current syntax file contains as "reserved words" the "return", "when", "forever", "block" and many other function names also i should mention that Emacs haskell mode, afaik, already does it. you can try to borrow their experience?

I have purposely left some items behind, just because resources are limited and we need to focus on some more valuable features. Do you think the ones I selected are the most valuable ones?

i tried to sort them in the importance order and for me the first 4 are most important (i often need them in my development process): - background compilation with errors highlighting - integration with WinHugs - syntax highlighting, including highlighting of bracket's pair - ability to "fall in" definition of identifier under cursor

Would you mind to delve into some more detail on the topics above on the EclipseFP mailing list? I am particularly interested on these four (in order of priority):

- ability to "fall in" definition of identifier under cursor - automatic management of import lists - syntax highlighting - auto-indenting with user-tuned style

i've said about everything except for syntax highlighting. there is no much to say - editor i use has vim-like features, which means that it's really great in this area. may be there is some Eclipse package that does the same? one feature i forgot to mention is code navigation. The structured code navigation should, imho, include files in the project mirroring the directory structure and on the next level each file should be splitted to "sections". In the absence of classes splitting to sections is widely used for the Haskell source files. For example, GHC library sources contains the following section headings: -- ----------------------------------------------------------------------------- -- Handy IOErrors -- ----------------------------------------------------------------------------- -- Handle Finalizers and so on. I personally adopted this headings style but it will be great to just allow to use regexp to define it. And inside each section one can see the identifiers it define Another, unstructured view should just list all identifiers in project and type in several chars from the identifier name to locate it in list. It's an explanation of this very helpful feature: my shell (FAR manager) saves all the commands i use (now this list contains ~5000 ones!) and when i open this list and start to type some word, it just filers whole list and show only commands which contains this sequence of letters. that is damn useful. just for example i open this window and typed "hugs" - FAR filters out 146 commands which includes this word: ---------------------------- Commands history *[hugs](146/5144) -------------------------- t C:\Base\Compiler\Hugs\runhugs.exe +st.qkoOuI -98 nocmt.hs

You can subscribe to our mailing list on http://lists.sourceforge.net/mailman/listinfo/eclipsefp-develop

done

...

Are you use GHC library?

Not yet, but it may turn to be a very wise decision to make in the future.

Simon said in his paper that many great features of Visual Haskell would be impossible without GHC running in background

EclipseFP is being written in Java, I wonder how the GHC library would be accessed on such a environment. Need to take a look at the paper.

http://haskell.org/haskellwiki/InterfacingOtherLanguages#Java -- Best regards, Bulat mailto:Bulat.Ziganshin@gmail.com

...

Are you use GHC library?

Not yet, but it may turn to be a very wise decision to make in the future. EclipseFP is being written in Java, I wonder how the GHC library would be accessed on such a environment. Need to take a look at the paper. I think the way to go is to call it via JNI. This has worked fine for me in the past on Windows with dlls, but not on Linux (because GHC did not yet support position-independent code at that time and I didn't manage to connect to a native library otherwise). Is this possible meanwhile?

Ciao, Leif

Cheers,

Thiago Arrais _______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

I have managed to call Haskell from Java on Linux -- I had to manually invoke gcc for the linking step because ghc won't output a shared object. It was very ugly indeed, but I can show you what I did if it will help. jvm-bridge also seems more developed now than when I was doing this, so maybe that's easier. keegan Thiago Arrais wrote:

Leif,

On 7/3/06, Leif Frenzel wrote:

...

I think the way to go is to call it via JNI. This has worked fine for me in the past on Windows with dlls

Although it required quite some glue code written in C, and it did not look very pretty. Have you already taken a look at this thing here

http://sourceforge.net/projects/jvm-bridge/

Seems like it will free us from writing much of the glue code.

I will definetely experiment with the GHC library + JVM-Bridge, it seems very much promising. Has anyone actually tried something like that?

Cheers,

Thiago Arrais _______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

Hi

The York Haskell Compiler (yhc) compiles to bytecode, and my memory suggests that there was an implementation of the required runtime system written in Java. My mail archive confirms this: Neil Mitchell wrote about it on haskell-cafe on March 28 under the topic of 'Haskell's Market'. It does indeed, but it would probably be faster to write a trivial C/Java wrapper round the C version of the runtime - still reasonably simple with Yhc. Yhc also compiles directly to .NET bytecodes, if the same was done with Java bytecodes it might be even easier than integrating an interpretter. Any way, there should be many ways to get Haskell, Yhc and Java all playing nicely.

My feeling is that such an approach might be a smoother (and more portable) integration than JNI. Of course, you'd need to be able compile GHC with YHC, which I cannot guess the feasability of. My guess is that would be very hard, I can only assume the GHC source code is far from Haskell 98. Maybe once Haskell' is done and Yhc supports that then you'll have more of a chance. It would probably be very good if Yhc could compile GHC then porting GHC would become trivial.

There is of course the option to integrate with the (still not started) Yhc API which will certainly compile with Yhc, rather than relying on GHC. Thanks Neil